Financial Fraud Detection using Synthetic Dataset: An Analysis of Machine Learning Algorithms

This paper addresses the challenges of fraud detection in monetary transactions through a data-driven approach. In financial management, it is critical to detect and avoid monetary fraud. Here, we utilize a PaySim database using aggregated real-world mobile financial provider data from multiple countries worldwide. We devise the Synthetic Minority Over-sampling Technique (SMOTE) with a 0.45 sampling strategy to solve the data imbalance problem. This would help improve the model's ability to detect fraudulent activities while maintaining the integrity of class distributions. Key findings indicate that fraudulent activities were primarily concentrated on cash-out and transfer transactions. We also present a clear pattern between fraudulent and non-fraudulent transactions. Further, we identify key relationships between variables, aiding in feature selection represented by a correlation heatmap. Similarly, we perform a temporal analysis of fraud trends to identify fluctuations across different transaction steps. We propose a method that utilizes machine learning algorithms known as XGBoost, LightGBM, and Logistic regression to identify fraud activities with near-perfect accuracy. Results show that LightGBM performs better than Logistic Regression, achieving near-perfect recall and AUC-ROC scores, demonstrating superior fraud detection capabilities. Among other techniques, the XGBoost model achieved a perfect accuracy of 100%, precision of 100%, recall of 100%, F1 score of 100%, and an AUC score of 100%. The performance metrics exhibited by LightGBM are exceptional: an accuracy of 100%, precision of 100%, recall of 100%, F1 score of 100%, and an AUC score of 99%. Additionally, we employ three key non-informative attributes (e.g., nameOrig, nameDest, and isFlaggedFraud) to improve model performance and minimize data leakage. This research establishes a new standard in fraud detection activities in monetary transaction cases. Different stakeholders (e.g., financial organizations, investors, business unit) will benefit from this research by minimizing fraud activities and finding the best management practices to increase customer satisfaction by getting early fraud alerts.